Wednesday, December 3 | 3:20 p.m.-3:30 p.m. | W7-SSCH07-3 | Room S501
AI can help identify malignant lung nodules earlier -- especially if it is trained to use additional data beyond nodule size and growth, researchers have found.
In this Wednesday afternoon session, presenter Sylvain Bodard, MD, PhD, of Massachusetts General Hospital in Boston, will share study findings that could lead to more timely clinical decision-making regarding lung cancer -- and thus translate to improved patient outcomes due to earlier treatment.
Early detection of lung nodules is crucial for effective lung cancer diagnosis and treatment, but conventional screening and risk stratification rely on nodule size and growth, the group explained. It sought to compare the performance of an AI/machine-learning-based CADe/CADx device to traditional growth metrics, malignancy scoring via Lung-RADS, and radiologist assessment for earlier and more accurate lung cancer detection.
The researchers trained the AI model on data from 10,872 patients taken from the National Lung Screening Trial (NLST) and the Lung Image Database Consortium (LIDC). They defined ground truth as six or more years of stability for benign lesions and histopathology for malignant ones.
Next, they evaluated the performance of the model using an NLST test cohort of 2,080 patients who underwent two low-dose CT scans: one at 12 to 24 months before diagnosis and another at zero to 12 months after. Four thoracic radiologists assessed a subset of the exams without applying Lung-RADS, and six assessed a different subset using Lung-RADS.
The team's primary result was that the AI model outperformed conventional lung nodule growth feature assessment, including volume doubling time and Lung-RADS categories.
"These findings challenge the current paradigm positioning volume doubling time as a primary malignancy indicator," the group concluded.
Check out this presentation for more details on the investigators' results.
